]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. | |
2a1d9b7f RD |
3 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. |
4 | * Copyright (c) 2004 Voltaire, Inc. All rights reserved. | |
1da177e4 LT |
5 | * |
6 | * This software is available to you under a choice of one of two | |
7 | * licenses. You may choose to be licensed under the terms of the GNU | |
8 | * General Public License (GPL) Version 2, available from the file | |
9 | * COPYING in the main directory of this source tree, or the | |
10 | * OpenIB.org BSD license below: | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or | |
13 | * without modification, are permitted provided that the following | |
14 | * conditions are met: | |
15 | * | |
16 | * - Redistributions of source code must retain the above | |
17 | * copyright notice, this list of conditions and the following | |
18 | * disclaimer. | |
19 | * | |
20 | * - Redistributions in binary form must reproduce the above | |
21 | * copyright notice, this list of conditions and the following | |
22 | * disclaimer in the documentation and/or other materials | |
23 | * provided with the distribution. | |
24 | * | |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
26 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
27 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
28 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
29 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
30 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
31 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
32 | * SOFTWARE. | |
1da177e4 LT |
33 | */ |
34 | ||
35 | #include <linux/skbuff.h> | |
36 | #include <linux/rtnetlink.h> | |
fec14d2f | 37 | #include <linux/moduleparam.h> |
1da177e4 LT |
38 | #include <linux/ip.h> |
39 | #include <linux/in.h> | |
40 | #include <linux/igmp.h> | |
41 | #include <linux/inetdevice.h> | |
42 | #include <linux/delay.h> | |
43 | #include <linux/completion.h> | |
5a0e3ad6 | 44 | #include <linux/slab.h> |
1da177e4 | 45 | |
14c85021 ACM |
46 | #include <net/dst.h> |
47 | ||
1da177e4 LT |
48 | #include "ipoib.h" |
49 | ||
50 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG | |
51 | static int mcast_debug_level; | |
52 | ||
53 | module_param(mcast_debug_level, int, 0644); | |
54 | MODULE_PARM_DESC(mcast_debug_level, | |
55 | "Enable multicast debug tracing if > 0"); | |
56 | #endif | |
57 | ||
1da177e4 LT |
58 | struct ipoib_mcast_iter { |
59 | struct net_device *dev; | |
60 | union ib_gid mgid; | |
61 | unsigned long created; | |
62 | unsigned int queuelen; | |
63 | unsigned int complete; | |
64 | unsigned int send_only; | |
65 | }; | |
66 | ||
69911416 | 67 | /* |
1c0453d6 | 68 | * This should be called with the priv->lock held |
69911416 DL |
69 | */ |
70 | static void __ipoib_mcast_schedule_join_thread(struct ipoib_dev_priv *priv, | |
71 | struct ipoib_mcast *mcast, | |
72 | bool delay) | |
73 | { | |
0e5544d9 | 74 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) |
69911416 DL |
75 | return; |
76 | ||
77 | /* | |
78 | * We will be scheduling *something*, so cancel whatever is | |
79 | * currently scheduled first | |
80 | */ | |
81 | cancel_delayed_work(&priv->mcast_task); | |
82 | if (mcast && delay) { | |
83 | /* | |
84 | * We had a failure and want to schedule a retry later | |
85 | */ | |
86 | mcast->backoff *= 2; | |
87 | if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS) | |
88 | mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS; | |
89 | mcast->delay_until = jiffies + (mcast->backoff * HZ); | |
90 | /* | |
91 | * Mark this mcast for its delay, but restart the | |
92 | * task immediately. The join task will make sure to | |
93 | * clear out all entries without delays, and then | |
94 | * schedule itself to run again when the earliest | |
95 | * delay expires | |
96 | */ | |
97 | queue_delayed_work(priv->wq, &priv->mcast_task, 0); | |
98 | } else if (delay) { | |
99 | /* | |
100 | * Special case of retrying after a failure to | |
101 | * allocate the broadcast multicast group, wait | |
102 | * 1 second and try again | |
103 | */ | |
104 | queue_delayed_work(priv->wq, &priv->mcast_task, HZ); | |
105 | } else | |
106 | queue_delayed_work(priv->wq, &priv->mcast_task, 0); | |
107 | } | |
108 | ||
5a0e81f6 | 109 | static void ipoib_mcast_free(struct ipoib_mcast *mcast) |
1da177e4 LT |
110 | { |
111 | struct net_device *dev = mcast->dev; | |
b36f170b | 112 | int tx_dropped = 0; |
1da177e4 | 113 | |
5b095d98 | 114 | ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n", |
fcace2fe | 115 | mcast->mcmember.mgid.raw); |
1da177e4 | 116 | |
b63b70d8 SP |
117 | /* remove all neigh connected to this mcast */ |
118 | ipoib_del_neighs_by_gid(dev, mcast->mcmember.mgid.raw); | |
1da177e4 | 119 | |
1da177e4 LT |
120 | if (mcast->ah) |
121 | ipoib_put_ah(mcast->ah); | |
122 | ||
b36f170b MT |
123 | while (!skb_queue_empty(&mcast->pkt_queue)) { |
124 | ++tx_dropped; | |
8c608a32 | 125 | dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); |
b36f170b MT |
126 | } |
127 | ||
943c246e | 128 | netif_tx_lock_bh(dev); |
de903512 | 129 | dev->stats.tx_dropped += tx_dropped; |
943c246e | 130 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
131 | |
132 | kfree(mcast); | |
133 | } | |
134 | ||
135 | static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev, | |
136 | int can_sleep) | |
137 | { | |
138 | struct ipoib_mcast *mcast; | |
139 | ||
de6eb66b | 140 | mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC); |
1da177e4 LT |
141 | if (!mcast) |
142 | return NULL; | |
143 | ||
1da177e4 LT |
144 | mcast->dev = dev; |
145 | mcast->created = jiffies; | |
69911416 | 146 | mcast->delay_until = jiffies; |
ce5b65cc | 147 | mcast->backoff = 1; |
1da177e4 LT |
148 | |
149 | INIT_LIST_HEAD(&mcast->list); | |
150 | INIT_LIST_HEAD(&mcast->neigh_list); | |
151 | skb_queue_head_init(&mcast->pkt_queue); | |
152 | ||
1da177e4 LT |
153 | return mcast; |
154 | } | |
155 | ||
432c55ff | 156 | static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid) |
1da177e4 LT |
157 | { |
158 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
159 | struct rb_node *n = priv->multicast_tree.rb_node; | |
160 | ||
161 | while (n) { | |
162 | struct ipoib_mcast *mcast; | |
163 | int ret; | |
164 | ||
165 | mcast = rb_entry(n, struct ipoib_mcast, rb_node); | |
166 | ||
37c22a77 | 167 | ret = memcmp(mgid, mcast->mcmember.mgid.raw, |
1da177e4 LT |
168 | sizeof (union ib_gid)); |
169 | if (ret < 0) | |
170 | n = n->rb_left; | |
171 | else if (ret > 0) | |
172 | n = n->rb_right; | |
173 | else | |
174 | return mcast; | |
175 | } | |
176 | ||
177 | return NULL; | |
178 | } | |
179 | ||
180 | static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast) | |
181 | { | |
182 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
183 | struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL; | |
184 | ||
185 | while (*n) { | |
186 | struct ipoib_mcast *tmcast; | |
187 | int ret; | |
188 | ||
189 | pn = *n; | |
190 | tmcast = rb_entry(pn, struct ipoib_mcast, rb_node); | |
191 | ||
192 | ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw, | |
193 | sizeof (union ib_gid)); | |
194 | if (ret < 0) | |
195 | n = &pn->rb_left; | |
196 | else if (ret > 0) | |
197 | n = &pn->rb_right; | |
198 | else | |
199 | return -EEXIST; | |
200 | } | |
201 | ||
202 | rb_link_node(&mcast->rb_node, pn, n); | |
203 | rb_insert_color(&mcast->rb_node, &priv->multicast_tree); | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
208 | static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast, | |
209 | struct ib_sa_mcmember_rec *mcmember) | |
210 | { | |
211 | struct net_device *dev = mcast->dev; | |
212 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
7343b231 | 213 | struct ipoib_ah *ah; |
1da177e4 | 214 | int ret; |
d0de1362 | 215 | int set_qkey = 0; |
1da177e4 LT |
216 | |
217 | mcast->mcmember = *mcmember; | |
218 | ||
bea1e22d PM |
219 | /* Set the multicast MTU and cached Q_Key before we attach if it's |
220 | * the broadcast group. | |
221 | */ | |
1da177e4 LT |
222 | if (!memcmp(mcast->mcmember.mgid.raw, priv->dev->broadcast + 4, |
223 | sizeof (union ib_gid))) { | |
e1d50dce JM |
224 | spin_lock_irq(&priv->lock); |
225 | if (!priv->broadcast) { | |
226 | spin_unlock_irq(&priv->lock); | |
227 | return -EAGAIN; | |
228 | } | |
3fd0605c ES |
229 | /*update priv member according to the new mcast*/ |
230 | priv->broadcast->mcmember.qkey = mcmember->qkey; | |
231 | priv->broadcast->mcmember.mtu = mcmember->mtu; | |
232 | priv->broadcast->mcmember.traffic_class = mcmember->traffic_class; | |
233 | priv->broadcast->mcmember.rate = mcmember->rate; | |
234 | priv->broadcast->mcmember.sl = mcmember->sl; | |
235 | priv->broadcast->mcmember.flow_label = mcmember->flow_label; | |
236 | priv->broadcast->mcmember.hop_limit = mcmember->hop_limit; | |
237 | /* assume if the admin and the mcast are the same both can be changed */ | |
238 | if (priv->mcast_mtu == priv->admin_mtu) | |
239 | priv->admin_mtu = | |
240 | priv->mcast_mtu = | |
241 | IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu)); | |
242 | else | |
243 | priv->mcast_mtu = | |
244 | IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu)); | |
245 | ||
1da177e4 | 246 | priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey); |
e1d50dce | 247 | spin_unlock_irq(&priv->lock); |
e622f2f4 | 248 | priv->tx_wr.remote_qkey = priv->qkey; |
d0de1362 | 249 | set_qkey = 1; |
1da177e4 LT |
250 | } |
251 | ||
252 | if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
253 | if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { | |
5b095d98 | 254 | ipoib_warn(priv, "multicast group %pI6 already attached\n", |
fcace2fe | 255 | mcast->mcmember.mgid.raw); |
1da177e4 LT |
256 | |
257 | return 0; | |
258 | } | |
259 | ||
260 | ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid), | |
d0de1362 | 261 | &mcast->mcmember.mgid, set_qkey); |
1da177e4 | 262 | if (ret < 0) { |
5b095d98 | 263 | ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n", |
fcace2fe | 264 | mcast->mcmember.mgid.raw); |
1da177e4 LT |
265 | |
266 | clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags); | |
267 | return ret; | |
268 | } | |
269 | } | |
270 | ||
271 | { | |
272 | struct ib_ah_attr av = { | |
273 | .dlid = be16_to_cpu(mcast->mcmember.mlid), | |
274 | .port_num = priv->port, | |
275 | .sl = mcast->mcmember.sl, | |
276 | .ah_flags = IB_AH_GRH, | |
bf6a9e31 | 277 | .static_rate = mcast->mcmember.rate, |
1da177e4 LT |
278 | .grh = { |
279 | .flow_label = be32_to_cpu(mcast->mcmember.flow_label), | |
280 | .hop_limit = mcast->mcmember.hop_limit, | |
281 | .sgid_index = 0, | |
282 | .traffic_class = mcast->mcmember.traffic_class | |
283 | } | |
284 | }; | |
1da177e4 LT |
285 | av.grh.dgid = mcast->mcmember.mgid; |
286 | ||
7343b231 | 287 | ah = ipoib_create_ah(dev, priv->pd, &av); |
3874397c MM |
288 | if (IS_ERR(ah)) { |
289 | ipoib_warn(priv, "ib_address_create failed %ld\n", | |
290 | -PTR_ERR(ah)); | |
291 | /* use original error */ | |
292 | return PTR_ERR(ah); | |
1da177e4 | 293 | } else { |
624d01f8 OG |
294 | spin_lock_irq(&priv->lock); |
295 | mcast->ah = ah; | |
296 | spin_unlock_irq(&priv->lock); | |
297 | ||
5b095d98 | 298 | ipoib_dbg_mcast(priv, "MGID %pI6 AV %p, LID 0x%04x, SL %d\n", |
fcace2fe | 299 | mcast->mcmember.mgid.raw, |
1da177e4 LT |
300 | mcast->ah->ah, |
301 | be16_to_cpu(mcast->mcmember.mlid), | |
302 | mcast->mcmember.sl); | |
303 | } | |
304 | } | |
305 | ||
306 | /* actually send any queued packets */ | |
943c246e | 307 | netif_tx_lock_bh(dev); |
1da177e4 LT |
308 | while (!skb_queue_empty(&mcast->pkt_queue)) { |
309 | struct sk_buff *skb = skb_dequeue(&mcast->pkt_queue); | |
69cce1d1 | 310 | |
943c246e | 311 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
312 | |
313 | skb->dev = dev; | |
1da177e4 LT |
314 | if (dev_queue_xmit(skb)) |
315 | ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n"); | |
936d7de3 | 316 | |
943c246e | 317 | netif_tx_lock_bh(dev); |
1da177e4 | 318 | } |
943c246e | 319 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
320 | |
321 | return 0; | |
322 | } | |
323 | ||
e8224e4b YE |
324 | void ipoib_mcast_carrier_on_task(struct work_struct *work) |
325 | { | |
326 | struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv, | |
327 | carrier_on_task); | |
5ee95120 | 328 | struct ib_port_attr attr; |
e8224e4b | 329 | |
5ee95120 MS |
330 | if (ib_query_port(priv->ca, priv->port, &attr) || |
331 | attr.state != IB_PORT_ACTIVE) { | |
332 | ipoib_dbg(priv, "Keeping carrier off until IB port is active\n"); | |
333 | return; | |
334 | } | |
335 | ||
894021a7 DL |
336 | /* |
337 | * Take rtnl_lock to avoid racing with ipoib_stop() and | |
338 | * turning the carrier back on while a device is being | |
339 | * removed. However, ipoib_stop() will attempt to flush | |
340 | * the workqueue while holding the rtnl lock, so loop | |
341 | * on trylock until either we get the lock or we see | |
342 | * FLAG_OPER_UP go away as that signals that we are bailing | |
343 | * and can safely ignore the carrier on work. | |
344 | */ | |
345 | while (!rtnl_trylock()) { | |
346 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) | |
347 | return; | |
348 | else | |
349 | msleep(20); | |
350 | } | |
c84ca6d2 DL |
351 | if (!ipoib_cm_admin_enabled(priv->dev)) |
352 | dev_set_mtu(priv->dev, min(priv->mcast_mtu, priv->admin_mtu)); | |
e8224e4b YE |
353 | netif_carrier_on(priv->dev); |
354 | rtnl_unlock(); | |
355 | } | |
356 | ||
faec2f7b SH |
357 | static int ipoib_mcast_join_complete(int status, |
358 | struct ib_sa_multicast *multicast) | |
1da177e4 | 359 | { |
faec2f7b | 360 | struct ipoib_mcast *mcast = multicast->context; |
1da177e4 LT |
361 | struct net_device *dev = mcast->dev; |
362 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
363 | ||
d2fe937c DL |
364 | ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n", |
365 | test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? | |
366 | "sendonly " : "", | |
fcace2fe | 367 | mcast->mcmember.mgid.raw, status); |
1da177e4 | 368 | |
faec2f7b | 369 | /* We trap for port events ourselves. */ |
e7a623d2 RD |
370 | if (status == -ENETRESET) { |
371 | status = 0; | |
a9c8ba58 | 372 | goto out; |
e7a623d2 | 373 | } |
faec2f7b SH |
374 | |
375 | if (!status) | |
376 | status = ipoib_mcast_join_finish(mcast, &multicast->rec); | |
377 | ||
378 | if (!status) { | |
ce5b65cc | 379 | mcast->backoff = 1; |
69911416 | 380 | mcast->delay_until = jiffies; |
55c9adde | 381 | |
e8224e4b | 382 | /* |
0b39578b | 383 | * Defer carrier on work to priv->wq to avoid a |
d2fe937c DL |
384 | * deadlock on rtnl_lock here. Requeue our multicast |
385 | * work too, which will end up happening right after | |
386 | * our carrier on task work and will allow us to | |
387 | * send out all of the non-broadcast joins | |
e8224e4b | 388 | */ |
d2fe937c | 389 | if (mcast == priv->broadcast) { |
1c0453d6 | 390 | spin_lock_irq(&priv->lock); |
0b39578b | 391 | queue_work(priv->wq, &priv->carrier_on_task); |
d2fe937c | 392 | __ipoib_mcast_schedule_join_thread(priv, NULL, 0); |
1c0453d6 | 393 | goto out_locked; |
d2fe937c | 394 | } |
69911416 | 395 | } else { |
d1178cbc JG |
396 | bool silent_fail = |
397 | test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) && | |
398 | status == -EINVAL; | |
399 | ||
400 | if (mcast->logcount < 20) { | |
401 | if (status == -ETIMEDOUT || status == -EAGAIN || | |
402 | silent_fail) { | |
d2fe937c DL |
403 | ipoib_dbg_mcast(priv, "%smulticast join failed for %pI6, status %d\n", |
404 | test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "", | |
69911416 DL |
405 | mcast->mcmember.mgid.raw, status); |
406 | } else { | |
d2fe937c DL |
407 | ipoib_warn(priv, "%smulticast join failed for %pI6, status %d\n", |
408 | test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ? "sendonly " : "", | |
69911416 DL |
409 | mcast->mcmember.mgid.raw, status); |
410 | } | |
d1178cbc JG |
411 | |
412 | if (!silent_fail) | |
413 | mcast->logcount++; | |
e7a623d2 | 414 | } |
e7a623d2 | 415 | |
d2fe937c DL |
416 | if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) && |
417 | mcast->backoff >= 2) { | |
418 | /* | |
419 | * We only retry sendonly joins once before we drop | |
420 | * the packet and quit trying to deal with the | |
421 | * group. However, we leave the group in the | |
422 | * mcast list as an unjoined group. If we want to | |
423 | * try joining again, we simply queue up a packet | |
424 | * and restart the join thread. The empty queue | |
425 | * is why the join thread ignores this group. | |
426 | */ | |
427 | mcast->backoff = 1; | |
428 | netif_tx_lock_bh(dev); | |
429 | while (!skb_queue_empty(&mcast->pkt_queue)) { | |
430 | ++dev->stats.tx_dropped; | |
431 | dev_kfree_skb_any(skb_dequeue(&mcast->pkt_queue)); | |
432 | } | |
433 | netif_tx_unlock_bh(dev); | |
1c0453d6 DL |
434 | } else { |
435 | spin_lock_irq(&priv->lock); | |
d2fe937c DL |
436 | /* Requeue this join task with a backoff delay */ |
437 | __ipoib_mcast_schedule_join_thread(priv, mcast, 1); | |
1c0453d6 DL |
438 | goto out_locked; |
439 | } | |
69911416 | 440 | } |
e7a623d2 | 441 | out: |
1c0453d6 DL |
442 | spin_lock_irq(&priv->lock); |
443 | out_locked: | |
444 | /* | |
445 | * Make sure to set mcast->mc before we clear the busy flag to avoid | |
446 | * racing with code that checks for BUSY before checking mcast->mc | |
447 | */ | |
69911416 DL |
448 | if (status) |
449 | mcast->mc = NULL; | |
1c0453d6 DL |
450 | else |
451 | mcast->mc = multicast; | |
452 | clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); | |
453 | spin_unlock_irq(&priv->lock); | |
e7a623d2 | 454 | complete(&mcast->done); |
1c0453d6 | 455 | |
faec2f7b | 456 | return status; |
1da177e4 LT |
457 | } |
458 | ||
08bc3276 AE |
459 | /* |
460 | * Caller must hold 'priv->lock' | |
461 | */ | |
462 | static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast) | |
1da177e4 LT |
463 | { |
464 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1c0453d6 | 465 | struct ib_sa_multicast *multicast; |
1da177e4 LT |
466 | struct ib_sa_mcmember_rec rec = { |
467 | .join_state = 1 | |
468 | }; | |
469 | ib_sa_comp_mask comp_mask; | |
470 | int ret = 0; | |
471 | ||
08bc3276 AE |
472 | if (!priv->broadcast || |
473 | !test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) | |
474 | return -EINVAL; | |
475 | ||
5b095d98 | 476 | ipoib_dbg_mcast(priv, "joining MGID %pI6\n", mcast->mcmember.mgid.raw); |
1da177e4 LT |
477 | |
478 | rec.mgid = mcast->mcmember.mgid; | |
479 | rec.port_gid = priv->local_gid; | |
97f52eb4 | 480 | rec.pkey = cpu_to_be16(priv->pkey); |
1da177e4 LT |
481 | |
482 | comp_mask = | |
483 | IB_SA_MCMEMBER_REC_MGID | | |
484 | IB_SA_MCMEMBER_REC_PORT_GID | | |
485 | IB_SA_MCMEMBER_REC_PKEY | | |
486 | IB_SA_MCMEMBER_REC_JOIN_STATE; | |
487 | ||
c3acdc06 DL |
488 | if (mcast != priv->broadcast) { |
489 | /* | |
490 | * RFC 4391: | |
491 | * The MGID MUST use the same P_Key, Q_Key, SL, MTU, | |
492 | * and HopLimit as those used in the broadcast-GID. The rest | |
493 | * of attributes SHOULD follow the values used in the | |
494 | * broadcast-GID as well. | |
495 | */ | |
1da177e4 | 496 | comp_mask |= |
d0df6d6d RD |
497 | IB_SA_MCMEMBER_REC_QKEY | |
498 | IB_SA_MCMEMBER_REC_MTU_SELECTOR | | |
499 | IB_SA_MCMEMBER_REC_MTU | | |
500 | IB_SA_MCMEMBER_REC_TRAFFIC_CLASS | | |
501 | IB_SA_MCMEMBER_REC_RATE_SELECTOR | | |
502 | IB_SA_MCMEMBER_REC_RATE | | |
503 | IB_SA_MCMEMBER_REC_SL | | |
504 | IB_SA_MCMEMBER_REC_FLOW_LABEL | | |
505 | IB_SA_MCMEMBER_REC_HOP_LIMIT; | |
1da177e4 LT |
506 | |
507 | rec.qkey = priv->broadcast->mcmember.qkey; | |
d0df6d6d RD |
508 | rec.mtu_selector = IB_SA_EQ; |
509 | rec.mtu = priv->broadcast->mcmember.mtu; | |
510 | rec.traffic_class = priv->broadcast->mcmember.traffic_class; | |
511 | rec.rate_selector = IB_SA_EQ; | |
512 | rec.rate = priv->broadcast->mcmember.rate; | |
1da177e4 LT |
513 | rec.sl = priv->broadcast->mcmember.sl; |
514 | rec.flow_label = priv->broadcast->mcmember.flow_label; | |
d0df6d6d | 515 | rec.hop_limit = priv->broadcast->mcmember.hop_limit; |
c3acdc06 DL |
516 | |
517 | /* | |
c3852ab0 DL |
518 | * Send-only IB Multicast joins do not work at the core |
519 | * IB layer yet, so we can't use them here. However, | |
520 | * we are emulating an Ethernet multicast send, which | |
521 | * does not require a multicast subscription and will | |
522 | * still send properly. The most appropriate thing to | |
523 | * do is to create the group if it doesn't exist as that | |
524 | * most closely emulates the behavior, from a user space | |
525 | * application perspecitive, of Ethernet multicast | |
526 | * operation. For now, we do a full join, maybe later | |
527 | * when the core IB layers support send only joins we | |
528 | * will use them. | |
c3acdc06 | 529 | */ |
c3852ab0 | 530 | #if 0 |
c3acdc06 DL |
531 | if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) |
532 | rec.join_state = 4; | |
533 | #endif | |
1da177e4 | 534 | } |
08bc3276 | 535 | spin_unlock_irq(&priv->lock); |
1da177e4 | 536 | |
1c0453d6 | 537 | multicast = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port, |
faec2f7b SH |
538 | &rec, comp_mask, GFP_KERNEL, |
539 | ipoib_mcast_join_complete, mcast); | |
08bc3276 | 540 | spin_lock_irq(&priv->lock); |
1c0453d6 DL |
541 | if (IS_ERR(multicast)) { |
542 | ret = PTR_ERR(multicast); | |
faec2f7b | 543 | ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret); |
1c0453d6 | 544 | /* Requeue this join task with a backoff delay */ |
69911416 | 545 | __ipoib_mcast_schedule_join_thread(priv, mcast, 1); |
1c0453d6 DL |
546 | clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); |
547 | spin_unlock_irq(&priv->lock); | |
69911416 | 548 | complete(&mcast->done); |
08bc3276 | 549 | spin_lock_irq(&priv->lock); |
faec2f7b | 550 | } |
08bc3276 | 551 | return 0; |
1da177e4 LT |
552 | } |
553 | ||
c4028958 | 554 | void ipoib_mcast_join_task(struct work_struct *work) |
1da177e4 | 555 | { |
c4028958 DH |
556 | struct ipoib_dev_priv *priv = |
557 | container_of(work, struct ipoib_dev_priv, mcast_task.work); | |
558 | struct net_device *dev = priv->dev; | |
94232d9c | 559 | struct ib_port_attr port_attr; |
69911416 DL |
560 | unsigned long delay_until = 0; |
561 | struct ipoib_mcast *mcast = NULL; | |
1da177e4 | 562 | |
0e5544d9 | 563 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) |
1da177e4 LT |
564 | return; |
565 | ||
94232d9c ES |
566 | if (ib_query_port(priv->ca, priv->port, &port_attr) || |
567 | port_attr.state != IB_PORT_ACTIVE) { | |
568 | ipoib_dbg(priv, "port state is not ACTIVE (state = %d) suspending join task\n", | |
569 | port_attr.state); | |
570 | return; | |
571 | } | |
68f9d83c | 572 | priv->local_lid = port_attr.lid; |
94232d9c | 573 | |
55ee3ab2 | 574 | if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid, NULL)) |
24bd1e4e | 575 | ipoib_warn(priv, "ib_query_gid() failed\n"); |
1da177e4 LT |
576 | else |
577 | memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid)); | |
578 | ||
69911416 DL |
579 | spin_lock_irq(&priv->lock); |
580 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) | |
581 | goto out; | |
582 | ||
1da177e4 | 583 | if (!priv->broadcast) { |
20b83382 RD |
584 | struct ipoib_mcast *broadcast; |
585 | ||
69911416 | 586 | broadcast = ipoib_mcast_alloc(dev, 0); |
20b83382 | 587 | if (!broadcast) { |
1da177e4 | 588 | ipoib_warn(priv, "failed to allocate broadcast group\n"); |
69911416 DL |
589 | /* |
590 | * Restart us after a 1 second delay to retry | |
591 | * creating our broadcast group and attaching to | |
592 | * it. Until this succeeds, this ipoib dev is | |
593 | * completely stalled (multicast wise). | |
594 | */ | |
595 | __ipoib_mcast_schedule_join_thread(priv, NULL, 1); | |
596 | goto out; | |
1da177e4 LT |
597 | } |
598 | ||
20b83382 | 599 | memcpy(broadcast->mcmember.mgid.raw, priv->dev->broadcast + 4, |
1da177e4 | 600 | sizeof (union ib_gid)); |
20b83382 | 601 | priv->broadcast = broadcast; |
1da177e4 | 602 | |
1da177e4 | 603 | __ipoib_mcast_add(dev, priv->broadcast); |
1da177e4 LT |
604 | } |
605 | ||
606 | if (!test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { | |
69911416 DL |
607 | if (IS_ERR_OR_NULL(priv->broadcast->mc) && |
608 | !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags)) { | |
609 | mcast = priv->broadcast; | |
69911416 DL |
610 | if (mcast->backoff > 1 && |
611 | time_before(jiffies, mcast->delay_until)) { | |
612 | delay_until = mcast->delay_until; | |
613 | mcast = NULL; | |
614 | } | |
615 | } | |
616 | goto out; | |
1da177e4 LT |
617 | } |
618 | ||
69911416 DL |
619 | /* |
620 | * We'll never get here until the broadcast group is both allocated | |
621 | * and attached | |
622 | */ | |
623 | list_for_each_entry(mcast, &priv->multicast_list, list) { | |
624 | if (IS_ERR_OR_NULL(mcast->mc) && | |
625 | !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags) && | |
d2fe937c DL |
626 | (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) || |
627 | !skb_queue_empty(&mcast->pkt_queue))) { | |
69911416 | 628 | if (mcast->backoff == 1 || |
d2fe937c | 629 | time_after_eq(jiffies, mcast->delay_until)) { |
1da177e4 | 630 | /* Found the next unjoined group */ |
d2fe937c DL |
631 | init_completion(&mcast->done); |
632 | set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); | |
08bc3276 AE |
633 | if (ipoib_mcast_join(dev, mcast)) { |
634 | spin_unlock_irq(&priv->lock); | |
635 | return; | |
636 | } | |
d2fe937c | 637 | } else if (!delay_until || |
69911416 DL |
638 | time_before(mcast->delay_until, delay_until)) |
639 | delay_until = mcast->delay_until; | |
1da177e4 | 640 | } |
1da177e4 LT |
641 | } |
642 | ||
d2fe937c DL |
643 | mcast = NULL; |
644 | ipoib_dbg_mcast(priv, "successfully started all multicast joins\n"); | |
1da177e4 | 645 | |
69911416 | 646 | out: |
d2fe937c DL |
647 | if (delay_until) { |
648 | cancel_delayed_work(&priv->mcast_task); | |
649 | queue_delayed_work(priv->wq, &priv->mcast_task, | |
650 | delay_until - jiffies); | |
651 | } | |
69911416 DL |
652 | if (mcast) { |
653 | init_completion(&mcast->done); | |
654 | set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags); | |
08bc3276 | 655 | ipoib_mcast_join(dev, mcast); |
69911416 DL |
656 | } |
657 | spin_unlock_irq(&priv->lock); | |
1da177e4 LT |
658 | } |
659 | ||
660 | int ipoib_mcast_start_thread(struct net_device *dev) | |
661 | { | |
662 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1c0453d6 | 663 | unsigned long flags; |
1da177e4 LT |
664 | |
665 | ipoib_dbg_mcast(priv, "starting multicast thread\n"); | |
666 | ||
1c0453d6 | 667 | spin_lock_irqsave(&priv->lock, flags); |
69911416 | 668 | __ipoib_mcast_schedule_join_thread(priv, NULL, 0); |
1c0453d6 | 669 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
670 | |
671 | return 0; | |
672 | } | |
673 | ||
efc82eee | 674 | int ipoib_mcast_stop_thread(struct net_device *dev) |
1da177e4 LT |
675 | { |
676 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1c0453d6 | 677 | unsigned long flags; |
1da177e4 LT |
678 | |
679 | ipoib_dbg_mcast(priv, "stopping multicast thread\n"); | |
680 | ||
1c0453d6 | 681 | spin_lock_irqsave(&priv->lock, flags); |
1da177e4 | 682 | cancel_delayed_work(&priv->mcast_task); |
1c0453d6 | 683 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 | 684 | |
efc82eee | 685 | flush_workqueue(priv->wq); |
1da177e4 | 686 | |
1da177e4 LT |
687 | return 0; |
688 | } | |
689 | ||
5a0e81f6 | 690 | static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast) |
1da177e4 LT |
691 | { |
692 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1da177e4 LT |
693 | int ret = 0; |
694 | ||
e07832b6 | 695 | if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) |
69911416 DL |
696 | ipoib_warn(priv, "ipoib_mcast_leave on an in-flight join\n"); |
697 | ||
698 | if (!IS_ERR_OR_NULL(mcast->mc)) | |
e07832b6 SH |
699 | ib_sa_free_multicast(mcast->mc); |
700 | ||
faec2f7b | 701 | if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) { |
5b095d98 | 702 | ipoib_dbg_mcast(priv, "leaving MGID %pI6\n", |
fcace2fe | 703 | mcast->mcmember.mgid.raw); |
1da177e4 | 704 | |
faec2f7b | 705 | /* Remove ourselves from the multicast group */ |
9eae554c RD |
706 | ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid, |
707 | be16_to_cpu(mcast->mcmember.mlid)); | |
faec2f7b | 708 | if (ret) |
9eae554c | 709 | ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret); |
69911416 DL |
710 | } else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) |
711 | ipoib_dbg(priv, "leaving with no mcmember but not a " | |
712 | "SENDONLY join\n"); | |
1da177e4 | 713 | |
1da177e4 LT |
714 | return 0; |
715 | } | |
716 | ||
432c55ff CL |
717 | /* |
718 | * Check if the multicast group is sendonly. If so remove it from the maps | |
719 | * and add to the remove list | |
720 | */ | |
721 | void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid, | |
722 | struct list_head *remove_list) | |
723 | { | |
724 | /* Is this multicast ? */ | |
725 | if (*mgid == 0xff) { | |
726 | struct ipoib_mcast *mcast = __ipoib_mcast_find(priv->dev, mgid); | |
727 | ||
728 | if (mcast && test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
729 | list_del(&mcast->list); | |
730 | rb_erase(&mcast->rb_node, &priv->multicast_tree); | |
731 | list_add_tail(&mcast->list, remove_list); | |
732 | } | |
733 | } | |
734 | } | |
735 | ||
50be28de | 736 | void ipoib_mcast_remove_list(struct list_head *remove_list) |
5a0e81f6 CL |
737 | { |
738 | struct ipoib_mcast *mcast, *tmcast; | |
739 | ||
740 | list_for_each_entry_safe(mcast, tmcast, remove_list, list) { | |
50be28de | 741 | ipoib_mcast_leave(mcast->dev, mcast); |
5a0e81f6 CL |
742 | ipoib_mcast_free(mcast); |
743 | } | |
744 | } | |
745 | ||
b63b70d8 | 746 | void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb) |
1da177e4 LT |
747 | { |
748 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
749 | struct ipoib_mcast *mcast; | |
943c246e | 750 | unsigned long flags; |
b63b70d8 | 751 | void *mgid = daddr + 4; |
700db99d | 752 | |
943c246e | 753 | spin_lock_irqsave(&priv->lock, flags); |
1da177e4 | 754 | |
b3e2749b | 755 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags) || |
20b83382 RD |
756 | !priv->broadcast || |
757 | !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) { | |
de903512 | 758 | ++dev->stats.tx_dropped; |
479a0796 MT |
759 | dev_kfree_skb_any(skb); |
760 | goto unlock; | |
761 | } | |
762 | ||
1da177e4 | 763 | mcast = __ipoib_mcast_find(dev, mgid); |
d2fe937c | 764 | if (!mcast || !mcast->ah) { |
1da177e4 | 765 | if (!mcast) { |
d2fe937c DL |
766 | /* Let's create a new send only group now */ |
767 | ipoib_dbg_mcast(priv, "setting up send only multicast group for %pI6\n", | |
768 | mgid); | |
769 | ||
770 | mcast = ipoib_mcast_alloc(dev, 0); | |
771 | if (!mcast) { | |
772 | ipoib_warn(priv, "unable to allocate memory " | |
773 | "for multicast structure\n"); | |
774 | ++dev->stats.tx_dropped; | |
775 | dev_kfree_skb_any(skb); | |
776 | goto unlock; | |
777 | } | |
1da177e4 | 778 | |
d2fe937c DL |
779 | set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags); |
780 | memcpy(mcast->mcmember.mgid.raw, mgid, | |
781 | sizeof (union ib_gid)); | |
782 | __ipoib_mcast_add(dev, mcast); | |
783 | list_add_tail(&mcast->list, &priv->multicast_list); | |
784 | } | |
1da177e4 LT |
785 | if (skb_queue_len(&mcast->pkt_queue) < IPOIB_MAX_MCAST_QUEUE) |
786 | skb_queue_tail(&mcast->pkt_queue, skb); | |
b36f170b | 787 | else { |
de903512 | 788 | ++dev->stats.tx_dropped; |
1da177e4 | 789 | dev_kfree_skb_any(skb); |
b36f170b | 790 | } |
d2fe937c DL |
791 | if (!test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) { |
792 | __ipoib_mcast_schedule_join_thread(priv, NULL, 0); | |
793 | } | |
794 | } else { | |
b63b70d8 SP |
795 | struct ipoib_neigh *neigh; |
796 | ||
797 | spin_unlock_irqrestore(&priv->lock, flags); | |
798 | neigh = ipoib_neigh_get(dev, daddr); | |
799 | spin_lock_irqsave(&priv->lock, flags); | |
800 | if (!neigh) { | |
b63b70d8 | 801 | neigh = ipoib_neigh_alloc(daddr, dev); |
b63b70d8 SP |
802 | if (neigh) { |
803 | kref_get(&mcast->ah->ref); | |
804 | neigh->ah = mcast->ah; | |
805 | list_add_tail(&neigh->list, &mcast->neigh_list); | |
1da177e4 LT |
806 | } |
807 | } | |
721d67cd | 808 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 | 809 | ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN); |
b63b70d8 SP |
810 | if (neigh) |
811 | ipoib_neigh_put(neigh); | |
721d67cd | 812 | return; |
1da177e4 LT |
813 | } |
814 | ||
479a0796 | 815 | unlock: |
943c246e | 816 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
817 | } |
818 | ||
819 | void ipoib_mcast_dev_flush(struct net_device *dev) | |
820 | { | |
821 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
822 | LIST_HEAD(remove_list); | |
988bd503 | 823 | struct ipoib_mcast *mcast, *tmcast; |
1da177e4 LT |
824 | unsigned long flags; |
825 | ||
826 | ipoib_dbg_mcast(priv, "flushing multicast list\n"); | |
827 | ||
828 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 829 | |
988bd503 EC |
830 | list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { |
831 | list_del(&mcast->list); | |
832 | rb_erase(&mcast->rb_node, &priv->multicast_tree); | |
833 | list_add_tail(&mcast->list, &remove_list); | |
1da177e4 LT |
834 | } |
835 | ||
836 | if (priv->broadcast) { | |
3cd96564 | 837 | rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree); |
988bd503 EC |
838 | list_add_tail(&priv->broadcast->list, &remove_list); |
839 | priv->broadcast = NULL; | |
1da177e4 LT |
840 | } |
841 | ||
842 | spin_unlock_irqrestore(&priv->lock, flags); | |
843 | ||
69911416 DL |
844 | /* |
845 | * make sure the in-flight joins have finished before we attempt | |
846 | * to leave | |
847 | */ | |
a9c8ba58 | 848 | list_for_each_entry_safe(mcast, tmcast, &remove_list, list) |
69911416 | 849 | if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) |
a9c8ba58 ES |
850 | wait_for_completion(&mcast->done); |
851 | ||
50be28de | 852 | ipoib_mcast_remove_list(&remove_list); |
1da177e4 LT |
853 | } |
854 | ||
3e4aa12f | 855 | static int ipoib_mcast_addr_is_valid(const u8 *addr, const u8 *broadcast) |
5e47596b | 856 | { |
5e47596b JG |
857 | /* reserved QPN, prefix, scope */ |
858 | if (memcmp(addr, broadcast, 6)) | |
859 | return 0; | |
860 | /* signature lower, pkey */ | |
861 | if (memcmp(addr + 7, broadcast + 7, 3)) | |
862 | return 0; | |
863 | return 1; | |
864 | } | |
865 | ||
c4028958 | 866 | void ipoib_mcast_restart_task(struct work_struct *work) |
1da177e4 | 867 | { |
c4028958 DH |
868 | struct ipoib_dev_priv *priv = |
869 | container_of(work, struct ipoib_dev_priv, restart_task); | |
870 | struct net_device *dev = priv->dev; | |
22bedad3 | 871 | struct netdev_hw_addr *ha; |
1da177e4 LT |
872 | struct ipoib_mcast *mcast, *tmcast; |
873 | LIST_HEAD(remove_list); | |
874 | unsigned long flags; | |
335a64a5 | 875 | struct ib_sa_mcmember_rec rec; |
1da177e4 | 876 | |
69911416 DL |
877 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) |
878 | /* | |
879 | * shortcut...on shutdown flush is called next, just | |
880 | * let it do all the work | |
881 | */ | |
882 | return; | |
1da177e4 | 883 | |
69911416 | 884 | ipoib_dbg_mcast(priv, "restarting multicast task\n"); |
4e0ab200 | 885 | |
932ff279 | 886 | local_irq_save(flags); |
e308a5d8 | 887 | netif_addr_lock(dev); |
78bfe0b5 | 888 | spin_lock(&priv->lock); |
1da177e4 LT |
889 | |
890 | /* | |
891 | * Unfortunately, the networking core only gives us a list of all of | |
892 | * the multicast hardware addresses. We need to figure out which ones | |
893 | * are new and which ones have been removed | |
894 | */ | |
895 | ||
896 | /* Clear out the found flag */ | |
897 | list_for_each_entry(mcast, &priv->multicast_list, list) | |
898 | clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); | |
899 | ||
900 | /* Mark all of the entries that are found or don't exist */ | |
22bedad3 | 901 | netdev_for_each_mc_addr(ha, dev) { |
1da177e4 LT |
902 | union ib_gid mgid; |
903 | ||
22bedad3 | 904 | if (!ipoib_mcast_addr_is_valid(ha->addr, dev->broadcast)) |
5e47596b JG |
905 | continue; |
906 | ||
22bedad3 | 907 | memcpy(mgid.raw, ha->addr + 4, sizeof mgid); |
1da177e4 | 908 | |
1da177e4 LT |
909 | mcast = __ipoib_mcast_find(dev, &mgid); |
910 | if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
911 | struct ipoib_mcast *nmcast; | |
912 | ||
335a64a5 OG |
913 | /* ignore group which is directly joined by userspace */ |
914 | if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) && | |
915 | !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) { | |
5b095d98 | 916 | ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %pI6\n", |
fcace2fe | 917 | mgid.raw); |
335a64a5 OG |
918 | continue; |
919 | } | |
920 | ||
1da177e4 | 921 | /* Not found or send-only group, let's add a new entry */ |
5b095d98 | 922 | ipoib_dbg_mcast(priv, "adding multicast entry for mgid %pI6\n", |
fcace2fe | 923 | mgid.raw); |
1da177e4 LT |
924 | |
925 | nmcast = ipoib_mcast_alloc(dev, 0); | |
926 | if (!nmcast) { | |
927 | ipoib_warn(priv, "unable to allocate memory for multicast structure\n"); | |
928 | continue; | |
929 | } | |
930 | ||
931 | set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags); | |
932 | ||
933 | nmcast->mcmember.mgid = mgid; | |
934 | ||
935 | if (mcast) { | |
936 | /* Destroy the send only entry */ | |
179e0917 | 937 | list_move_tail(&mcast->list, &remove_list); |
1da177e4 LT |
938 | |
939 | rb_replace_node(&mcast->rb_node, | |
940 | &nmcast->rb_node, | |
941 | &priv->multicast_tree); | |
942 | } else | |
943 | __ipoib_mcast_add(dev, nmcast); | |
944 | ||
945 | list_add_tail(&nmcast->list, &priv->multicast_list); | |
946 | } | |
947 | ||
948 | if (mcast) | |
949 | set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags); | |
950 | } | |
951 | ||
952 | /* Remove all of the entries don't exist anymore */ | |
953 | list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) { | |
954 | if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) && | |
955 | !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) { | |
5b095d98 | 956 | ipoib_dbg_mcast(priv, "deleting multicast group %pI6\n", |
fcace2fe | 957 | mcast->mcmember.mgid.raw); |
1da177e4 LT |
958 | |
959 | rb_erase(&mcast->rb_node, &priv->multicast_tree); | |
960 | ||
961 | /* Move to the remove list */ | |
179e0917 | 962 | list_move_tail(&mcast->list, &remove_list); |
1da177e4 LT |
963 | } |
964 | } | |
78bfe0b5 MT |
965 | |
966 | spin_unlock(&priv->lock); | |
e308a5d8 | 967 | netif_addr_unlock(dev); |
932ff279 | 968 | local_irq_restore(flags); |
1da177e4 | 969 | |
69911416 DL |
970 | /* |
971 | * make sure the in-flight joins have finished before we attempt | |
972 | * to leave | |
973 | */ | |
974 | list_for_each_entry_safe(mcast, tmcast, &remove_list, list) | |
975 | if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) | |
976 | wait_for_completion(&mcast->done); | |
977 | ||
50be28de | 978 | ipoib_mcast_remove_list(&remove_list); |
962121b4 | 979 | |
69911416 DL |
980 | /* |
981 | * Double check that we are still up | |
982 | */ | |
983 | if (test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) { | |
984 | spin_lock_irqsave(&priv->lock, flags); | |
985 | __ipoib_mcast_schedule_join_thread(priv, NULL, 0); | |
986 | spin_unlock_irqrestore(&priv->lock, flags); | |
987 | } | |
1da177e4 LT |
988 | } |
989 | ||
8ae5a8a2 RD |
990 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG |
991 | ||
1da177e4 LT |
992 | struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev) |
993 | { | |
994 | struct ipoib_mcast_iter *iter; | |
995 | ||
996 | iter = kmalloc(sizeof *iter, GFP_KERNEL); | |
997 | if (!iter) | |
998 | return NULL; | |
999 | ||
1000 | iter->dev = dev; | |
1732b0ef | 1001 | memset(iter->mgid.raw, 0, 16); |
1da177e4 LT |
1002 | |
1003 | if (ipoib_mcast_iter_next(iter)) { | |
1732b0ef | 1004 | kfree(iter); |
1da177e4 LT |
1005 | return NULL; |
1006 | } | |
1007 | ||
1008 | return iter; | |
1009 | } | |
1010 | ||
1da177e4 LT |
1011 | int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter) |
1012 | { | |
1013 | struct ipoib_dev_priv *priv = netdev_priv(iter->dev); | |
1014 | struct rb_node *n; | |
1015 | struct ipoib_mcast *mcast; | |
1016 | int ret = 1; | |
1017 | ||
1018 | spin_lock_irq(&priv->lock); | |
1019 | ||
1020 | n = rb_first(&priv->multicast_tree); | |
1021 | ||
1022 | while (n) { | |
1023 | mcast = rb_entry(n, struct ipoib_mcast, rb_node); | |
1024 | ||
1025 | if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw, | |
1026 | sizeof (union ib_gid)) < 0) { | |
1027 | iter->mgid = mcast->mcmember.mgid; | |
1028 | iter->created = mcast->created; | |
1029 | iter->queuelen = skb_queue_len(&mcast->pkt_queue); | |
1030 | iter->complete = !!mcast->ah; | |
1031 | iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY)); | |
1032 | ||
1033 | ret = 0; | |
1034 | ||
1035 | break; | |
1036 | } | |
1037 | ||
1038 | n = rb_next(n); | |
1039 | } | |
1040 | ||
1041 | spin_unlock_irq(&priv->lock); | |
1042 | ||
1043 | return ret; | |
1044 | } | |
1045 | ||
1046 | void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter, | |
1047 | union ib_gid *mgid, | |
1048 | unsigned long *created, | |
1049 | unsigned int *queuelen, | |
1050 | unsigned int *complete, | |
1051 | unsigned int *send_only) | |
1052 | { | |
1053 | *mgid = iter->mgid; | |
1054 | *created = iter->created; | |
1055 | *queuelen = iter->queuelen; | |
1056 | *complete = iter->complete; | |
1057 | *send_only = iter->send_only; | |
1058 | } | |
8ae5a8a2 RD |
1059 | |
1060 | #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ |